Rainfall generator for the Meuse basin: Description of simulations with and without a memory term and uncertainty analysis

The rainfall generator has been developed to generate long synthetic sequences of daily precipitation and temperature for the Meuse basin [see e.g. Leander and Buishand (2004), Leander et al. (2005), and Buishand and Leander (2011)] using the nearest-neighbour resampling (NNR) technique. These sequences have been used for discharge simulations with the semi-distributed HBV model (Aalders et al., 2004; Leander et al., 2005) to estimate the design discharge for flood protection works in the Netherlands. For the non-tidal embanked part of the river, the 1250-year return level of the discharge at Borgharen (near the city of Maastricht) is presently used as design discharge. An important source of uncertainty of this design discharge is the length of the historical records used for resampling. In order to study the sensitivity of the 1250-year return level to the choice of the historical data, several 20,000-year simulations were conducted with various 33-year subsets of the 1930-1998 period as well as a 20,000-year simulation based on the whole 1930-1998 period (Leander and Buishand, 2008). Apart from the average winter rainfall of the subset, it turned out that the presence or absence of the year 1995 in the subset strongly influenced the estimate of the 1250-year return level of the maximum 10-day winter basin-average rainfall as well as the estimated 1250-year return level of the discharge at Borgharen (Kramer et al., 2008). A difficulty with this sensitivity analysis is that it does not provide the standard deviation of the estimated return level. Therefore, for the Rhine basin a jackknife method was used to determine the uncertainty of the return level (Schmeits et al., 2014). In the present report this jackknife method is applied for the rainfall generator for the Meuse basin. Further, two alternative forms of the rainfall generator are considered to downweight the influence of the year 1995 on the estimated return levels, and the maximum 4-, 10- and 20-day basin-average rainfall amounts are studied rather than the maximum 10-day basin-average rainfall only.